This project aims to create an open-source platform for deploying ML algorithms for modeling and design of Analog/Mixed-Signal circuits. BWRC researchers have created a Python-based tool (Berkeley Analog Generator) that leverages that Google/Skywater open process design kit to enable analog/mixed-signal circuit design and layout generation. We have also developed several machine learning frameworks (based on graph neural networks and deep reinforcement learning) for analog/mixed-signal circuit design and modeling.
In this project, we plan to create an open-source platform in Google Cloud that leverages BAG as the execution engine for circuit layout generation, and connects it to our existing machine-learning frameworks (e.g. ACnet: https://arxiv.org/abs/2203.15913) for circuit design while also leaving the programming hooks to instantiate other algorithms.
